libmagic: ship the complete identification database

[brandonpayton]

Purpose

Make libmagic a first-class Kandelo Homebrew dependency with the complete upstream type database and native in-process support for zlib, bzip2, and xz payloads.

The database is compiled by a version-matched native FILE_COMPILE helper. Its on-disk format is fixed-width and endian-tagged, so the formula removes the earlier host-pointer-size truncation assumption instead of preserving a partial 14k-line database.

Formula

• builds and installs static libmagic for wasm32;
• generates the complete 5.45 magic.mgc from all 342 upstream source files;
• installs the source magic files as provenance;
• enables built-in zlib, bzip2, and xz decoding through declared tap dependencies;
• publishes static pkg-config linkage with dependency keg paths and libm;
• embeds the stable Homebrew opt path for default magic_load discovery;
• leaves the file CLI to a separate consumer PR.

Existing ABI-18 evidence

The artifact-producing formula body is unchanged from the previously validated ABI-18 build:

• isolated Homebrew source install and literal brew test --keep-tmp --verbose libmagic passed;
• Node classified WebAssembly, PDF, zlib, bzip2, and xz inputs;
• Chromium BrowserKernel classified the same inputs with PATH=/no-such-path;
• database assertions passed for 342 source files, 45,082 logical source lines, 22,642 fixed-width records, and 8,513,392 bytes;
• WebAssembly recognition came from aggregated source line 42,046, beyond the former partial-database cutoff;
• the fork-instrumented test consumer manually disassembled to user ABI 18;
• strict Homebrew audit and style passed.

An independent source-only review subsequently replayed the ABI 18 runtime classification paths and rebuilt a libmagic side module against the current PIC bzip2/xz/zlib formula sources; both the classification and dlopen/dlsym consumer checks passed.

Source merge and bottle publication

Compression source PR #25 is merged at tap main a3ede334ad27aac7c4d5826bac761b9faf768f20. The declared bzip2, xz, and zlib source formulae now resolve from main, so this libmagic source formula no longer has a peer-branch dependency. This PR contains no bottle block, archive, receipt, sidecar, or package-registry artifact and may merge independently after fresh formula review.

Bottle construction and installability claims remain separate. Before publishing a libmagic bottle:

• sdk: reserve an 8 MiB main-thread shadow stack for linked programs kandelo#842, #873, #874, #881, #882, #885, and #890 must land;
• trusted ABI-18 bzip2, xz, and zlib bottles must be published first;
• the trusted publisher must build from the final libmagic tap head, record Homebrew receipts/provenance, and produce the bottle hash from those exact bytes;
• stock-guest Node and Chromium acceptance must pour libmagic and its dependencies from Homebrew bottles, with no source-build fallback.

Tap #22 is relevant only if a branch-based pre-merge dry run is needed. Automattic/kandelo#867 is merged and is not a gate.

Restack and current-main merge evidence

The two libmagic-owned commits were replayed onto tap main a3ede334ad27aac7c4d5826bac761b9faf768f20, and a third commit changes only the support import to the canonical installed Tap.fetch path. README conflict resolution was additive. Before merge, an independent review refreshed exact tap main aa99c67c632c4f5f0f6ac56850c2b325151e0b78; git merge-tree --write-tree succeeded cleanly, adding only Formula/libmagic.rb and one README line, with the reviewed formula blob unchanged.

Current head: 3784d6bd89e391f8e9660d90180f197e7c51ce2f.

Validation through Kandelo's scripts/dev-shell.sh passed:

• Homebrew Ruby syntax for the formula and shared support;
• shared support suite: 21 runs, 186 assertions;
• brew style Formula/libmagic.rb;
• literal Homebrew formula loading from a tap path, resolving pkgconf, automattic/kandelo-homebrew/bzip2, xz, and zlib;
• branch ancestry, three-commit ownership, range-diff inspection, a clean merge tree against current main, git diff --check, and a clean worktree;
• a main-relative diff containing only Formula/libmagic.rb and one additive README line;
• no changes to Formula/bzip2.rb or Formula/xz.rb, and no bottle metadata.

The old validated formula blob was 5710428d9c4e3e5065e5f493af99027c6fbd69c3; the new import-adjusted blob is 0ef72dfadb31c034174e66da97e70f819779e214. Every line after the support import is byte-identical, with SHA-256 255ee00f07aecc3bea393ff8cc1b523f87f9155b3dd4f94f538e09300c31d94e.

This PR changes only tap formula source and documentation. It does not change Kandelo's kernel, host runtime, SDK, or ABI.

[brandonpayton]

Independent devil review found that the original BSD-2-Clause declaration understated the licensing of the installed 5.45 database. Commit 6f3c169 corrects it to the complete Homebrew expression: BSD-2-Clause-Darwin + BSD-2-Clause + public domain.

Revalidated after the correction:

• clean-prefix brew audit --strict automattic/kandelo-homebrew/libmagic
• clean-prefix brew style automattic/kandelo-homebrew/libmagic
• git diff --check

The reviewer found no runtime, database-completeness, static-linking, fork-instrumentation, or ABI issue. The stacked file CLI adds the suggested separate gzip-header runtime coverage.

[brandonpayton]

Independent devil's-advocate source-only review at 3784d6bd89e391f8e9660d90180f197e7c51ce2f: ACCEPT.

The review refreshed exact tap main aa99c67c632c4f5f0f6ac56850c2b325151e0b78; the clean merge tree adds only Formula/libmagic.rb and one README line, with the reviewed formula blob unchanged. It rechecked the official file 5.45 source and composite license, complete 342-fragment database, native generator format, stable opt lookup path, direct compression dependencies, static pkg-config interface, and installed Tap.fetch loader.

Fresh dev-shell syntax, style, literal formula-load, support, dependency, diff, and no-artifact checks passed. A fresh ABI 18 replay classified Wasm/PDF and zlib/bzip2/xz payloads without external commands. A libmagic side module built against the current PIC bzip2/xz/zlib sources also passed dlopen/dlsym. The diff contains no bottle, sidecar, archive, receipt, provenance, JSON, Wasm, or duplicated dependency formula. No source-formula blocker remains.